Skip to main content
Vectorsoft

Why GitHub Copilot is our favorite harness — for now

Adam H. •

AIGitHub CopilotAgentic WorkflowDevelopment

For the moment, GitHub Copilot is our favorite AI coding harness.

Claude
Codex
GitHub Copilot
Three harnesses. Our pick today: Copilot.

It gives us model choice, good value, tight GitHub integration, and a UI we enjoy using. It also covers the development features we rely on. In a field moving this quickly, that makes it our pick today. We reserve the right to change our minds.

By harness, we mean everything around the AI model that helps it do useful work: the interface, tools, permissions, and environments where it runs code. Those choices shape how easily you can give an agent a task, guide it, and review the result.

Model choice, with surprisingly good value

Copilot lets us choose models from providers including Anthropic, OpenAI, and Google without changing the rest of our workflow. We can pick a model to suit the task and keep working in the same environment. Access varies by plan and feature; GitHub’s model comparison lists the options.

The pricing still makes us wonder how they pull this off.

Part of the value comes from extra usage bundled into the subscription. Paid individual plans include a variable flex allotment above their base credits. Paid plans also get a 10% discount on model costs with automatic model selection. Those benefits have conditions, so the savings depend on how you work. GitHub’s billing documentation explains the details.

We do not know the commercial arrangements behind it. From our side, access to several model providers in one reasonably priced tool is an appealing deal.

GitHub integration ties the work together

For a team already working in GitHub, the integration is a substantial advantage. The Copilot app brings issues, branches, pull requests, reviews, and automated check results into the same place as agent sessions. You can follow a task from the issue through implementation and review. GitHub’s app documentation describes the workflow.

That connection matters. When an agent produces a change, we need to know what it changed, whether the checks passed, and whether it solved the original problem. Keeping the request, the work, and the review together makes that easier.

The development features we need are here

Copilot covers almost everything we reach for in Claude Code and Codex, including their command-line workflows. In our day-to-day development, we have not found a meaningful gap. The capabilities we care about include:

  • Cloud environments. Run agent sessions in cloud sandboxes or local workspaces through the Copilot app.
  • A CLI and mobile access. Work in the terminal, then monitor and steer an interactive session from GitHub.com or GitHub Mobile. Steering requires remote control to be enabled; session syncing alone gives you a view of the work.
  • Built-in browser tools. Copilot in VS Code can inspect and interact with a running application through the editor’s integrated browser.
  • Computer use. Copilot CLI and the Copilot app can interact with desktop applications on Windows and macOS. Computer use is in public preview, must be enabled, and can be restricted by administrators.

Exact feature parity varies by plan, app, operating system, and organization policy. For the development work we do, the coverage is there.

We like the UI

We simply like using it. That deserves a place on the list for a tool we spend so much time in.

As agents take on longer tasks, more of our job becomes guiding their work and reviewing the output. Copilot’s interface makes that process easy to follow. We can stay oriented while the agent works, then focus on the changes it produced.

Where it falls short for us

GitHub Copilot is not yet our general work tool in the way ChatGPT/Codex and Claude are. For research, documents, presentations, and broader business work, we still reach for those other environments.

Copilot’s chat, integrations, and computer use already let it work beyond code. Its focus, though, remains software development. Some people will prefer that separation.

Microsoft seems to be making a similar distinction. Microsoft 365 Copilot connects to organizational information and apps such as Word, Excel, Outlook, and Teams. Our reading is that Microsoft sees that offering as its broader enterprise work tool, with GitHub Copilot focused on developers.

We also miss having a browser extension. When the work is on a page you already have open, an assistant alongside it can be extremely helpful. Claude in Chrome, for example, can read and interact with websites from the browser side panel.

As of this writing, we have not found an equivalent first-party GitHub Copilot extension for everyday browsing. Its integrated browser tools serve a different workflow. This is one reason we keep more than one assistant available.

What does it cost?

Published monthly prices in USD, checked October 8, 2026. Paid individual totals include the current flex allotment, which can change.

Plan Monthly subscription Included monthly AI credits
Free $0 Limited allowance
Pro $10 1,500 ($15 of usage)
Pro+ $39 7,000 ($70 of usage)
Max $100 20,000 ($200 of usage)
Business $19 per seat 1,900 per seat ($19), pooled
Enterprise $39 per seat 3,900 per seat ($39), pooled

Business adds centralized management and policy controls. Enterprise requires GitHub Enterprise Cloud. Verified students can use the free Student plan; eligible teachers and maintainers of popular open-source projects can receive Pro free. See GitHub’s plans and eligibility.

One AI credit represents $0.01 of usage. Model choice and token consumption determine how long the allowance lasts. Credits do not roll over, and additional usage needs a budget. Code completions and next-edit suggestions remain unlimited on paid plans; agent usage is metered. See GitHub’s billing details.

Allow for infrastructure costs too: Copilot code review consumes GitHub Actions minutes as well as AI credits, as described in GitHub’s billing update.

For individuals, the choice comes down to model access and usage. Teams also need to weigh shared budgets and administrative controls. We would compare a representative month of actual work before deciding which plan offers the best value.

Today, Copilot is the harness we prefer for building software. The broader work tools and browser extensions still give us reasons to use other assistants. We are comfortable with that: our favorite tool only needs to keep earning its place.